CariGAN: Stanford graduate student Kaidi Cao will join fellow AI researchers Jing Liao from City University of Hong Kong and Microsoft's Lu Yuan at SIGGRAPH Asia to be held in Tokyo in December, to present an incredible neural network that caricatures design.
This particular AI actually consists of two genetically adversarial networks (GAN), and is called CariGAN. The first of its neural networks, CariGeoGAN, determines the geometry of a person in a photo and renders it as a caricature. CariStyGAN, the other half of CariGAN, does the "metaphor of style" or applies the artistic look to the design.
To enrich CariGAN with the ability to turn a boring photo into a very enjoyable design, the system was trained in thousands of human-generated images.
To determine the effectiveness of the machine, the researchers conducted two studies. The first was to ensure that the AI caricatures retained the identity of the portrait. and the second study to determine the overall effectiveness of the "design" compared to human parts.
Both surveys seem to have been quite successful:
CariGANs AI can also analyze frames from a video and create caricatures. The following images are created by the frames of a video that President Trump talks about.
Images from TNW